Consent, enforced at execution.
QODIQA is a deterministic runtime consent enforcement standard for AI systems. It operates at the inference boundary, where consent becomes a precondition to execution rather than a declaration.
normative and informative documents in the corpus
enforcement control points in the full framework
core control points for minimal conformance
invariant principles, without exception
of specification, versioned and append-only
The gap is not regulatory. It is architectural.
Every action is intercepted at the execution boundary. Consent state is verified deterministically before any operation proceeds. No existing framework operates at that point. QODIQA defines and occupies it.
- Fail-closed: no valid consent, no execution
- Append-only audit with cryptographic verification
- Identical input, identical decision, every time
Illustrative enforcement trace. Request IDs and identities are examples.
Six invariants. No exceptions.
They govern deterministic consent enforcement across every action type, consent state and operational boundary.
Deterministic Enforcement
Given identical consent state and action request, the enforcer produces identical output. No probabilistic paths. No ambiguity.
Runtime Interception
Consent is verified at the moment of execution, not at ingestion or configuration time. The enforcement point is the action boundary.
Auditable by Construction
Every enforcement decision produces a cryptographically anchored audit record. Evidence is generated, not reconstructed.
Fail-Closed Default
In the absence of verified consent, execution is denied. The safe state is inaction. Uncertainty does not default to permission.
Consent Immutability
A recorded consent state cannot be retroactively altered. Withdrawal is prospective. History is append-only.
Scope Boundedness
Consent granted for one purpose does not extend to another. Every action maps to a declared scope, verified at runtime.
Frameworks define consent. None enforce it at runtime.
Any system that acts on human data. At any scale.
Systems that generate, infer, or route outputs derived from human data require deterministic consent enforcement at the execution boundary. It is a structural prerequisite for verifiable compliance.
Large-Scale Inference Platforms
Every inference event on multi-tenant infrastructure constitutes a discrete consent boundary. Enforcement liability cannot be delegated to the application layer or post-hoc audit.
Foundation Model Operators
Models trained on human data and deployed at scale carry consent obligations from training through every downstream inference event. Without runtime enforcement, consent integrity cannot be demonstrated at any stage.
AI Assistant Systems
Conversational and agentic systems that access, retrieve, or derive personal data on behalf of users require verifiable consent enforcement at every action boundary, not at session initiation alone.
Healthcare and Clinical AI
Special category health data access is scope-bound under GDPR Art. 9 and HIPAA. Runtime scope mismatch constitutes a violation regardless of prior documentation or organisational policy.
Financial and Risk Systems
Credit, underwriting, and algorithmic scoring decisions require auditable consent chains. Post-hoc reconstruction of consent state does not satisfy the technical burden of proof in regulatory proceedings.
Identity and Biometric Infrastructure
Biometric inference carries maximum regulatory exposure across all jurisdictions. Unverified, unscoped, or expired consent at execution constitutes an enforcement failure, not a process deviation.
Four certainties. Not risks.
Without runtime enforcement, audit, compliance, governance, and accountability operate on unverified assumptions. The system does not fail occasionally. It fails continuously.
Compliance investigations reconstruct consent state from records not designed as evidence. Reconstructive audit cannot establish execution-boundary facts, only approximations.
Consent in a CRM or policy document is a record of intent, not execution. Without cryptographic enforcement binding at runtime, no verification mechanism exists.
Systems without runtime enforcement can assert compliance, not demonstrate it. Regulatory proceedings require technical evidence of enforcement; documentation of intent does not satisfy that burden.
Scope mismatches, expired consent, and withdrawn permissions violate at the moment of inference. Without boundary enforcement, violations are invisible until post-hoc review. Damage precedes detection.
Complete. Normative. Open.
Structured for verifiability, designed for auditability, enforceable at runtime. Every document is versioned under QODIQA-STD-1.0.
An open, vendor-neutral standard for the AI execution boundary.
- Specification
- QODIQA-STD-1.0
- License
- QODIQA Public License · open · vendor-neutral
- Author & Architect
- Bogdan Duțescu